Metformin add‐on vs. antipsychotic switch vs. continued antipsychotic treatment plus healthy lifestyle education in overweight or obese youth with severe mental illness: results from the IMPACT trial

Abstract

Antipsychotics are used for many psychiatric conditions in youth. Although developmentally inappropriate weight gain and metabolic abnormalities, which are risk factors for premature cardiovascular mortality, are especially frequent in youth, optimal strategies to reduce pediatric antipsychotic‐induced overweight/obesity are unclear. The Improving Metabolic Parameters in Antipsychotic Child Treatment (IMPACT) was a randomized, parallel group, 24‐week clinical trial which enrolled overweight/obese, psychiatrically stable youth, aged 8‐19 years, with a DSM‐IV diagnosis of severe mental illness (schizophrenia spectrum disorder, bipolar spectrum disorder or psychotic depression), at four US universities. All of them had developed substantial weight gain following treatment with a second‐generation antipsychotic. The centralized, computer‐based randomization system assigned participants to unmasked treatment groups: metformin (MET); antipsychotic switch (aripiprazole or, if already exposed to that drug, perphenazine or molindone; SWITCH); or continued baseline antipsychotic (CONTROL). All participants received healthy lifestyle education. The primary outcome was body mass index (BMI) z‐score change from baseline, analyzed using estimated least squares means. Altogether, 127 participants were randomized: 49 to MET, 31 to SWITCH, and 47 to CONTROL. BMI z‐score decreased significantly with MET (week 24: –0.09±0.03, p=0.002) and SWITCH (week 24: –0.11±0.04, p=0.003), while it increased non‐significantly with CONTROL (week 24: +0.04±0.03). On 3‐way comparison, BMI z‐score changes differed significantly (p=0.001). MET and SWITCH were each superior to CONTROL (p=0.002), with effect sizes of 0.68 and 0.81 respectively, while MET and SWITCH did not differ. More gastrointestinal problems occurred in MET than in SWITCH or CONTROL. The data safety monitoring board closed the perphenazine‐SWITCH arm because 35.2% of subjects discontinued treatment due to psychiatric worsening. These data suggest that pediatric antipsychotic‐related overweight/obesity can be reduced by adding metformin or switching to a lower risk antipsychotic. Healthy lifestyle education is not sufficient to prevent ongoing BMI z‐score increase.

Antipsychotics are commonly used to treat many different mental disorders1 and are frequently associated with weight gain and metabolic abnormalities2, 3, 4, particularly in children and adolescents4, 5, 6, 7, which increases the risk for premature mortality8. Cardiometabolic risk monitoring in antipsychotic‐treated patients is often inadequate9, 10, especially in youth11, 12, 13. Interventions for antipsychotic‐related weight gain and metabolic abnormalities are still insufficiently established and, especially, infrequently implemented14, 15.

The Harvard Growth Study found that being overweight in adolescence is a more significant predictor of morbidity from coronary heart disease than being overweight as an adult16. A large population cohort study reported that adult coronary heart disease was positively associated with body mass index (BMI) at age 7‐13 for boys and 11‐13 for girls, with the risk increasing across the entire BMI distribution17. So, overweight/obesity induced by antipsychotic treatment in youth is a major public health concern.

Several strategies to reduce antipsychotic‐induced weight gain have been tested in adults14. They include behavioral lifestyle interventions18, switch to a lower‐risk antipsychotic19, and addition of topiramate20 or metformin21.

In adults, results of behavioral interventions have been mixed. A large randomized controlled trial (RCT) failed to show significant benefit22. A recent meta‐analysis18 of 41 RCTs (N=4,267) showed that, in adults with severe mental illness, individualized healthy lifestyle interventions lasting on average 22 weeks were able to reduce BMI by 0.63 kg/m² compared to control groups. However, after an average of 32 weeks post‐intervention, the effect size remained similar in 17 RCTs, but was no longer significant. Most studies had very low or low quality of evidence, and the statistically significant effects were considered very likely not to be clinically significant18.

In contrast to these data available in adults, there are almost no RCTs of behavioral weight reduction programs for children and adolescents with antipsychotic‐related weight gain. The only pediatric RCT of a 52‐week behavioral weight counseling intervention in adolescents with schizophrenia or bipolar disorder failed to demonstrate significant benefits23.

Among the pharmacological weight loss interventions for adults with severe mental illness, metformin is so far the best studied14, 21. In a meta‐analysis of 19 RCTs (N=1,279), the addition of metformin to antipsychotic treatment for an average of 3‐4 months significantly reduced body weight relative to control conditions, with a medium effect size of 0.6121.

The mechanism for weight loss induced by metformin is not entirely clear, but data suggest a variety of effects24. It has been well documented that metformin decreases hepatic gluconeogenesis and improves insulin sensitivity in the liver and muscle. Because insulin levels are elevated as part of insulin resistance following non‐physiologic weight gain, and insulin increases appetite, improvement of insulin resistance by metformin could reduce appetite and caloric intake. Additionally, metformin has been shown to affect hypothalamic signaling, regulating leptin sensitivity, gastrointestinal physiology and circadian rhythms, which may not only influence food intake, but also fat oxidation and fat storage in liver, skeletal muscle, and adipose tissue24.

Data on pharmacological interventions aimed at weight reduction in youth with antipsychotic‐induced overweight/obesity are far more limited than in adults. Only three RCTs of metfor‐min are available, lasting 12 to 16 weeks25, 26, 27. In a study of 39 youth with mixed psychiatric disorders, metformin separated from placebo on anthropometric, but not metabolic parameters25. In a trial of 49 youth with schizophrenia spectrum disorders, differences favoring metformin treatment were not statistically significant for body weight parameters, and no trends toward metabolic benefits were evident26. In a study of 60 youth with autism, metformin separated from placebo on anthropometric, but not on metabolic measures27.

Moreover, no pediatric or adult RCT to date has directly compared the effects of antipsychotic switching versus add‐on of a weight loss agent, and no pediatric trial has examined combined medication and behavioral treatment.

The objective of this study was to compare the efficacy and tolerability of the addition of metformin, the switch to an antipsychotic with lower risk for weight gain, and continued antipsychotic treatment, against the background of healthy lifestyle education (HLE), in youth with severe mental illness and clinically significant antipsychotic‐induced weight gain.

METHODS

The Improving Metabolic Parameters in Antipsychotic Child Treatment (IMPACT) was a randomized, unmasked parallel group clinical trial, approved and monitored by the Institutional Review Boards at Zucker/Hillside Hospital, Johns Hopkins University, University of Maryland, and University of North Carolina at Chapel Hill28. It was funded by the US National Institute of Mental Health (NIMH).

Participants

Youth aged 8‐19  years were enrolled for the study if they had a primary DSM‐IV diagnosis of schizophrenia spectrum disorder, bipolar spectrum disorder, or major depression with psychotic features. Psychiatric diagnoses were established at the screening appointment using the Leibenluft modification of the Kiddie Schedule for Affective Disorders and Schizophrenia for School‐Age Children ‐ Present and Lifetime version (K‐SADS‐PL)29, 30.

Further inclusion criteria were: a) having been treated with a second‐generation antipsychotic (SGA) – aripiprazole, asenapine, iloperidone, lurasidone, olanzapine, paliperidone, quetiapine, risperidone or ziprasidone – with a stable dose for at least 30 days; b) having been clinically stable on current treatment regimen, as assessed by Clinical Global Impression ‐ Severity and Improvement (CGI‐I and CGI‐S)31, for at least 30 days; c) having a BMI ≥85th percentile for age and gender (i.e., being overweight or obese); d) having had a substantial weight gain (>10% baseline weight) while taking the SGA; e) having a primary caretaker (parent(s), close relative functioning in loco parentis, legal guardian, or foster parent) who had known the person for at least 6 months before study entry; f) being able to participate in all aspects of the protocol per investigator clinical judgment.

Exclusion criteria were: a) treatment with more than one antipsychotic medication, or more than three total psychiatric medications (four were permitted if two were for attention‐deficit/hyperactivity disorder (ADHD)); b) antipsychotic treatment with clozapine (which is exclusively used for treatment‐refractory illness); c) psychiatric medication or dosage change within the past 30 days; d) any medication affecting glucose, insulin or lipid levels; d) any major neurological or medical illness affecting body weight or physical activity; e) abnormal fasting glucose (≥126 mg/dL) or serum creatinine (>1.3 mg/dL); f) substance dependence disorder (except tobacco dependence) in the past month; g) current or lifetime diagnosis or anorexia or bulimia nervosa; h) IQ <55; i) known hypersensitivity to aripiprazole, perphenazine or metformin; j) prior trials with aripiprazole or perphenazine lasting more than 2 weeks and stopped because of efficacy or tolerability concerns; k) significant risk of dangerousness to self or other; l) for female participants, pregnant, nursing or sexually active and unwilling to comply with double method contraceptive. ADHD medications and valproate were permitted.

Procedures

All 18‐19 year‐old participants and at least one parent of minors provided written informed consent; all participants <18 years old provided assent. Eligibility was determined based on a 3‐week screening period to establish psychiatric and physical health status and stability.

All eligible youth received HLE and were randomized to 24 weeks of open‐label treatment with either: a) add‐on metformin (MET); b) switch of SGA to a lower cardiometabolic risk antipsychotic (aripiprazole or, if previously exposed to this drug, molindone – prior to its removal from the US market – or perphenazine) (SWITCH); c) continued treatment with the current SGA (CONTROL). All metformin‐treated youth were given a daily multi‐vitamin, to prevent vitamin B12 deficiency32.

Molindone was originally chosen for the SWITCH condition when patients presented with significant weight gain while being on aripiprazole. It was selected as it produced the least weight gain in the Treatment of Early‐Onset Schizophrenia Spectrum Disorders (TEOSS)33 study. When it became unavailable in the US, the NIMH strongly recommended retaining the comparison between a SGA (aripiprazole) and a first‐generation antipsychotic (FGA) in the SWITCH arm. Among available FGAs, we chose perphenazine, as it had a better profile in terms of weight gain and metabolic changes than SGAs in the Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) project34.

Randomizations were performed through a centralized, computer‐based system. They were stratified for current SGA (risperidone, aripiprazole or “other antipsychotic”) and diagnosis (schizophrenia spectrum or other disorder). Patients, caregivers and study team members learned about the randomization assignment at the end of the baseline visit.

All study conditions involved 10 in‐person visits (at 0, 1, 2, 4, 6, 8, 12, 16, 20 and 24 weeks) and 6 phone sessions (at 3, 5, 7, 9, 10 and 11 weeks). Metabolic assessments were performed at baseline, 12 weeks and 24 weeks.

Psychiatric symptomatology was assessed at baseline and at weeks 12 and 24 by independent evaluators (physicians with ≥2 years of experience working with psychiatrically ill youth), blind to study condition and medication adverse events, using the 21‐item Brief Psychiatric Rating Scale for Children (BPRS‐C)35.

Ongoing monitoring during the treatment phase included review of psychiatric symptoms, adverse events and medication adherence by board‐certified child and adolescent psychiatrists. Review of psychiatric symptoms was done using the CGI‐I and CGI‐S. Adverse events were assessed by the Systematic Longitudinal Adverse Event Scale (SLAES)36, the Simpson‐Angus Extrapyramidal Symptoms Scale (SAEPS)37, the Barnes Akathisia Rating Scale (BARS)38 and the Abnormal Involuntary Movement Scale (AIMS)39.

All metabolic laboratory values at the main visits (baseline, 12 weeks, and 24 weeks) were obtained after an overnight fast of at least 8 hours. Glucose, insulin, triglycerides, total cholesterol, high‐density lipoprotein (HDL) cholesterol, low‐density lipoprotein (LDL) cholesterol, homeostatic model assessment for insulin resistance (HOMA‐IR), hemoglobin A1c (HgbA1c), and C‐reactive protein were measured.

For all cases, parent/patient self‐report adherence data were obtained regarding all psychotropic medications. For all participants receiving study medications (i.e., switch antipsychotic or metformin), information was also collected by pill count at all appointments. For all patients, antipsychotic blood levels were measured at weeks 12 and 24.

Patients could be withdrawn from the study and referred for clinical care based on either: a) participant or guardian request, b) judgment of the independent pediatrician or consulting pediatric endocrinologist that metabolic problems required treatment outside of the study, or c) significant worsening of psychiatric symptoms operationalized as a CGI‐I rating of “much worse” or “very much worse” on two successive occasions over a period of ≥2 weeks.

Youth might also be withdrawn if moderate/severe adverse events occurred that could not be addressed by dose adjustment or addition of permitted concomitant four medications or, if requested by the site principal investigator, due to non‐compliance or clinical status.

Treatment conditions

HLE consisted of education about strategies to enhance nutrition and physical activity, as well as regular weight monitoring40. The frequency of visits increased if participants gained 5% and 10% of baseline body weight.

For youth weighing <50 kg, the metformin titration started with 250 mg taken with dinner, which was increased by 250 mg (taken at breakfast) after 1 week, and subsequently by 250 mg steps on a weekly basis (taken twice daily, with breakfast and dinner), until 500 mg twice daily was reached. For youth weighing 50‐70 kg, the maximum metformin dose was 500 mg in the morning and 1,000 mg in the evening.

For youth weighing >70 kg, the metformin titration started with 500 mg taken with dinner, which was increased by 500 mg (taken at breakfast) after 1 week, and subsequently by 500 mg steps on a weekly basis (taken twice daily, with breakfast and dinner), until 1,000 mg twice daily was reached.

During the plateau cross‐titration from the baseline SGA to either aripiprazole or perphenazine, the baseline SGA was kept at the same dose for 3 weeks, and then tapered by 25% of the baseline dose over the next 3 weeks. Aripiprazole was initiated at 2 mg/day for one week, increased to 5 mg/day at week 2, and then increased in 5 mg steps to a maximum of 30 mg/day. Perphenazine was initiated at 4 mg/day and increased weekly by 4 mg to a maximum of 64 mg/day.

Both the metformin titration speed and SGA cross‐titration could be altered based on clinical response. In perphenazine‐SWITCH, benztropine 0.5 mg bid was required when perphenazine reached >8 mg/day. Metformin‐XR could be used if intolerable gastrointestinal problems occurred.

Outcomes

The primary outcome was BMI z‐score change. BMI z‐scores were calculated using the program provided by the Children's Nutrition Research Center at Baylor College of Medicine. Secondary outcomes included changes in other anthropometric measures, glucose and lipid parameters, C‐reactive protein, and the child‐rated Impact of Weight on Quality of Life‐Kids (IWQOL‐Kids)41.

Safety outcomes included longitudinal review of adverse effects elicited using the SLAES, SAEPS, BARS and AIMS, and psychiatric assessment by blinded raters using the CGI‐I and BPRS‐C.

Statistical analyses

Efficacy outcomes from subjects with baseline and ≥1 post‐baseline value were analyzed using a longitudinal mixed model entering predictors for treatment, visit (treated as a categorical variable with an unstructured covariance pattern to reflect correlation between each subject's visits), treatment‐by‐visit interaction, and baseline score into the model. Least squares means (LSMs) for change from baseline were estimated at each visit for each treatment group and for differences between each pair of treatment groups. LSMs are only reported from weeks 12 and 24, but models were fit using data from all available time points. All hypothesis‐testing analyses used a hierarchical approach to preserve power and eliminate inflation of the overall experiment‐wise error rate.

Imputation due to early termination was limited to variables collected only at week 12 and week 24. Discontinuation data from weeks 1‐11 was carried‐forward to week 12; data from weeks 13‐23 to week 24. Kenward‐Roger degrees of freedom42 were used in the denominator of significance tests to correct for multiple comparisons.

Time to discontinuation was estimated using Kaplan‐Meier survival curve with log‐rank tests to compare treatment groups. Demographics, adverse events, and other safety variables were summarized using basic descriptive statistics. Exploratory post‐hoc chi‐squared analyses compared the three groups for categorical weight gain and incidence of adverse events without multiple comparisons corrections.

Based on a power calculation, a sample size of 44 per group (total N=132) yielded 80% power to detect a significant difference regarding the primary outcome, BMI z‐score.

RESULTS

Participants

Between October 2009 and October 2013, 127 subjects were randomized (CONTROL=47; MET=49; SWITCH=31). Adverse event analyses excluded five participants (CONTROL=2, MET=2, SWITCH=1), who discontinued without providing adverse event information after learning their randomization assignment. Primary efficacy analyses included 121 participants (CONTROL=44; MET=47; SWITCH=30: aripiprazole=12, perphenazine=17; molindone =1) with ≥1 post‐baseline vital sign measurement.

Baseline characteristics

Patients’ baseline demographic and clinical characteristics are summarized in Table 1. Mean age was 13.7±3.3 years, 64.6% were male, and 52.7% were White. Primary diagnoses were bipolar spectrum disorder in 84.2% of patients, schizophrenia spectrum disorder in 9.4% and psychotic depression in 6.3%. The most common comorbid diagnoses were ADHD (35.4%), autism spectrum disorder (26.0%), anxiety disorders (25.2%), and oppositional defiant disorder or cosnduct disorder (21.2%). Among the participants, 52.0% had had prior psychiatric hospitalizations.

Table 1.

Demographic and clinical characteristics of the study participants at baseline (randomized population)

	Control (N=47)	Metformin (N=49)	Switch (N=31)	Total (N=127)
Age (mean±SD)	13.4±3.2	13.4±3.2	14.7±3.3	13.7±3.3
8‐12 years (%)	51.1	44.9	29.0	43.3
13‐17 years (%)	36.2	46.9	54.8	44.9
18+ years (%)	12.8	8.2	16.1	11.8
Gender (% males)	63.8	63.3	67.7	64.6
Ethnicity (%)
White	53.2	53.1	51.6	52.7
Black	27.6	28.6	29.0	28.3
Other	19.1	18.4	19.3	18.9
Primary psychiatric diagnosis (%)
Schizophrenia spectrum disorder	10.6	12.2	3.2	9.4
Bipolar spectrum disorder	83.0	83.7	22.6	84.2
Psychotic depression	6.4	4.1	9.7	6.3
Main comorbid diagnoses (%)
ADHD	31.9	34.7	41.9	35.4
Autism spectrum disorder	29.8	28.6	16.1	26.0
Anxiety disorders	29.8	28.6	12.9	25.2
Oppositional defiant disorder/conduct disorder	19.1	20.4	25.8	21.2
Antipsychotic medication at baseline (%)
Aripiprazole	51.1	40.8	48.4	46.4
Risperidone	36.2	40.8	38.7	38.6
Ziprasidone	8.5	8.2	6.4	7.9
Olanzapine	2.1	6.1	3.2	3.9
Quetiapine	2.1	4.1	3.2	3.1
Psychiatric co‐medication at baseline (%)
Psychostimulant	40.4	44.9	41.9	42.5
Antidepressant	34.0	36.7	38.7	36.2
Mood stabilizer	19.1	26.5	12.9	20.5
Non‐stimulant anti‐ADHD medication	31.9	4.1	12.9	16.5
Hypnotic/anxiolytic	12.8	18.4	19.3	16.5
Anticholinergic	10.6	8.2	6.4	8.7
Current antipsychotic use (months, mean±SD)	22.7±18.5	21.6±23.2	19.8±18.4	21.6±20.4
Total antipsychotic use (months, mean±SD)	30.8±20.9	29.0±26.7	30.0±20.3	29.9±23.1
Treated with multiple consecutive antipsychotics (%)	40.4	61.2	51.6	43.3
Prior psychiatric hospitalizations (%)	44.7	53.1	61.3	52.0

ADHD – attention‐deficit/hyperactivity disorder

Aripiprazole (46.4%) and risperidone (38.6%) were the most frequently antipsychotics received at baseline. Mean durations of current and total antipsychotic use were 21.6±20.4 and 29.9±23.1 months, respectively. Almost half (43.3%) had been treated consecutively with multiple antipsychotics. Treatment groups did not differ significantly on anthropometric, psychiatric and metabolic parameters at baseline.

Treatment

Most (72.3%) participants achieved their targeted MET dose by week 12. One additional participant achieved it by week 24. Mean endpoint metformin doses were 1,000±500 mg/day for youth weighing <50 kg; 1,250±500 mg/day for those weighing 50‐70 kg, and 1,766±442 mg/day for those weighing >70 kg.

Ten (83.3%) participants switching to aripiprazole did so by week 8; one was unable to completely discontinue his baseline SGA. Of those switching to perphenazine, 58.8% did so by week 8; 41.2% were unable to discontinue their baseline SGA. Mean endpoint doses were 12.8±9.0 mg/day for aripiprazole and 11.6±10.2 mg/day for perphenazine. Other psychotropics remained virtually unchanged throughout the study.

Mean treatment duration was 19.4±8.4 weeks for CONTROL, 20.3±7.2 weeks for MET, and 18.2±8.3 weeks for SWITCH (p=0.113), with marked discrepancy between aripiprazole (20.6±8.3 weeks), perphenazine (16.9±8.0 weeks) and molindone (12 weeks).

All‐cause discontinuation was significantly greater in perphenazine‐SWITCH (52.9%) than any other group (p=0.041; CONTROL=25.5%, MET=21.2%, SWITCH=36.6%, aripiprazole‐SWITCH=8.3%), primarily due to inadequate psychiatric efficacy (p=0.0014, CONTROL=6.3%, MET=4.1%, aripiprazole‐SWITCH=0%, perphenazine‐SWITCH=35.2%). Consequently, the NIMH data safety monitoring board closed the perphenazine‐SWITCH arm on February 8, 2013.

Primary outcome

Table 2 shows estimated changes from baseline to endpoint, p values and effect sizes for within‐group changes, the 3‐way comparison, and, if this comparison was significant, pairwise‐group comparisons for all efficacy outcomes.

Table 2.

Change from baseline to week 12 and to week 24 in anthropometric, metabolic and psychiatric outcomes (efficacy population)

	Control, within group (N=44)			Metformin, within group (N=47)			Switch, within group (N=30)			3‐way comparison	Control vs. Metformin		Control vs. Switch		Metformin vs. Switch
	LSM (SE)	p	ES	LSM (SE)	p	ES	LSM (SE)	p	ES	p	p	ES	p	ES	p	ES
BMI z‐score
Week 12	0.03 (0.01)	NS	0.05	–0.03 (0.01)	0.019	–0.07	–0.05 (0.02)	0.008	–0.10	0.002	0.005	0.40	0.002	0.51	NS	0.11
Week 24	0.04 (0.03)	NS	0.08	–0.09 (0.03)	0.002	–0.18	–0.11 (0.04)	0.003	–0.23	0.001	0.002	0.68	0.002	0.81	NS	0.13
Fasting insulin (μU/mL)
Week 12	–15.3 (11.3)	NS	–0.15	–33.0 (10.3)	0.002	–0.33	–5.7 (14.2)	NS	–0.06	NS		0.31		–0.17		–0.48
Week 24	17.4 (20.1)	NS	0.17	–18.6 (16.4)	NS	–0.18	4.4 (23.0)	NS	0.04	NS		0.42		0.15		–0.27
Weight (lbs)
Week 12	4.3 (0.8)	<0.0001	0.07	0.3 (0.7)	NS	0.00	0.3 (0.9)	NS	0.01	0.0002	0.0002	0.67	0.001	0.66	NS	–0.01
Week 24	8.5 (1.5)	<0.001	0.14	–0.4 (1.4)	NS	–0.01	0.3 (1.9)	NS	0.01	<0.0001	<0.0001	0.99	0.0008	0.91	NS	–0.08
BMI percentile
Week 12	0.5 (0.2)	0.017	0.12	–0.1 (0.2)	NS	–0.03	–0.4 (0.2)	NS	–0.09	0.016	0.029	0.22	0.008	0.31	NS	0.10
Week 24	0.7 (0.4)	0.045	0.19	–0.5 (0.4)	NS	–0.13	–0.9 (0.5)	0.054	–0.23	0.01	0.015	0.40	0.006	0.54	NS	0.13
Fasting glucose (mg/dL)
Week 12	3.1 (1.8)	NS	0.27	–6.1 (1.5)	0.0002	–0.53	–2.8 (2.1)	NS	–0.24	0.001	0.0002	0.95	0.036	0.62	NS	–0.34
Week 24	0.9 (2.0)	NS	0.08	–0.3 (1.7)	NS	–0.03	0.7 (2.6)	NS	0.06	NS		0.12		0.02		–0.10
HOMA‐IR
Week 12	–0.29 (0.41)	NS	–0.08	–1.43 (0.36)	0.0002	–0.4	–0.31 (0.50)	NS	–0.09	NS		0.57		0.01		–0.56
Week 24	0.63 (0.82)	NS	0.17	–0.46 (0.65)	NS	–0.13	0.30 (0.91)	NS	0.08	NS		0.33		0.10		–0.23
HbA1c (%)
Week 12	0.00 (0.04)	NS	0.00	–0.05 (0.03)	NS	–0.15	–0.03 (0.04)	NS	–0.10	NS		0.23		0.15		–0.08
Week 24	0.02 (0.04)	NS	0.05	–0.09 (0.04)	0.028	–0.27	0.01 (0.06)	NS	0.04	NS		0.43		0.01		–0.42
Total cholesterol (mg/dL)
Week 12	3.5 (4.0)	NS	0.11	–6.6 (3.8)	NS	–0.20	–1.9 (4.7)	NS	–0.06	NS		0.41		0.22		–0.19
Week 24	5.3 (4.4)	NS	0.16	–1.2 (4.2)	NS	–0.04	–10.2 (5.7)	NS	–0.31	NS		0.25		0.59		0.34
HDL‐cholesterol (mg/dL)
Week 12	0.7 (1.6)	NS	0.06	1.0 (1.5)	NS	0.08	–2.5 (1.9)	NS	–0.20	NS		–0.02		0.33		0.35
Week 24	0.4 (1.1)	NS	0.04	–0.1 (1.1)	NS	–0.01	–3.9 (1.5)	0.009	–0.31	NS		0.09		0.65		0.57
LDL‐cholesterol (mg/dL)
Week 12	2.6 (3.6)	NS	0.10	–6.0 (3.3)	NS	–0.22	2.2 (4.2)	NS	0.08	NS		0.39		0.02		–0.37
Week 24	3.6 (4.1)	NS	0.13	–3.9 (3.9)	NS	–0.14	–8.3 (5.3)	NS	–0.30	NS		0.31		0.49		0.18
Triglycerides (mg/dL)
Week 12	–5.8 (6.6)	NS	–0.10	–7.2 (6.2)	NS	–0.12	–6.6 (7.7)	NS	–0.11	NS		0.03		0.02		–0.01
Week 24	0.2 (9.1)	NS	0.00	14.7 (8.7)	NS	0.25	16.6 (12.0)	NS	0.28	NS		–0.27		–0.30		–0.04
C‐reactive protein (mg/L)
Week 12	0.07 (0.75)	NS	0.02	–0.52 (0.69)	NS	–0.13	–0.63 (1.09)	NS	–0.15	NS		0.16		0.19		0.03
Week 24	0.87 (0.82)	NS	0.21	–0.86 (0.77)	NS	–0.21	–1.64 (1.26)	NS	–0.39	NS		0.46		0.66		0.21
CGI severity score
Last – baseline	0.28 (0.78)	NS		–0.04 (0.86)	NS		–0.12 (1.11)	NS		NS
CGI improvement score
Last	3.38 (1.03)			3.43 (1.15)			3.80 (1.35)			NS
BPRS‐C total score
Week 12	–3.1 (1.7)	NS	–0.22	–2.8 (1.6)	NS	–0.20	–6.5 (2.1)	0.002	–0.45	NS		–0.03		0.31		0.34
Week 24	–4.3 (1.7)	0.012	–0.30	–4.6 (1.6)	0.006	–0.32	–4.8 (2.2)	0.031	–0.34	NS		0.03		0.05		0.02
IWQOL‐Kids physical comfort
Week 12	0.7 (0.5)	NS	0.14	1.5 (0.5)	0.002	0.29	1.1 (0.7)	NS	0.21	NS		–0.25		–0.12		0.13
Week 24	0.2 (0.6)	NS	0.04	1.6 (0.5)	0.005	0.31	2.9 (0.8)	0.0003	0.55	0.026	NS	–0.41	0.008	–0.78		–0.37
IWQOL‐Kids body esteem
Week 12	2.8 (1.0)	0.004	0.27	3.4 (0.9)	0.0002	0.32	2.5 (1.2)	0.037	0.24	NS		–0.10		0.06		0.15
Week 24	2.5 (0.9)	0.009	0.24	4.1 (0.9)	<0.0001	0.39	5.6 (1.2)	<0.0001	0.54	NS		–0.30		–0.59		–0.29
IWQOL‐Kids social life
Week 12	0.5 (0.7)	NS	0.11	2.2 (0.6)	0.0008	0.44	–1.3 (0.9)	NS	–0.26	0.005	NS	–0.41	NS	0.45	0.001	0.85
Week 24	0.6 (0.6)	NS	0.12	1.4 (0.6)	0.013	0.28	1.9 (0.8)	0.019	0.37	NS		–0.23		–0.36		–0.32
IWQOL‐Kids family relations
Week 12	0.6 (0.6)	NS	0.15	1.0 (0.5)	NS	0.28	0.3 (0.7)	NS	0.08	NS		–0.14		0.08		0.22
Week 24	0.3 (0.5)	NS	0.08	0.4 (0.4)	NS	0.10	0.6 (0.6)	NS	0.15	NS		–0.03		–0.10		–0.07
IWQOL‐Kids total
Week 12	4.9 (1.9)	0.014	0.25	8.1 (1.8)	<0.0001	0.41	2.4 (2.4)	NS	0.12	NS		–0.28		0.21		0.48
Week 24	3.8 (1.9)	0.047	0.19	7.4 (1.7)	<0.0001	0.38	10.7 (2.4)	<0.0001	0.55	NS		–0.33		–0.65		–0.31

LSM – least squares mean, SE – standard error, ES – effect size, BMI – body mass index, HOMA‐IR – homeostatic model of insulin resistance, HbA1c – hemoglobin A1c, HDL – high‐density lipoprotein, LDL – low‐density lipoprotein, CGI – Clinical Global Impression, BPRS‐C – Brief Psychiatric Rating Scale for Children, IWQOL‐Kids – Impact of Weight on Quality of Life‐Kids

The BMI z‐score decreased significantly in both MET (week 12: –0.03±0.01, p=0.019; week 24: –0.09±0.03, p=0.002) and SWITCH (week 12: –0.05±0.02, p=0.008, week 24: –0.11±0.04, p=0.003), while it increased non‐significantly in CONTROL (week 12: +0.03±0.01, week 24: +0.04±0.03).

The BMI z‐score change differed significantly between the three groups at weeks 12 (p=0.002) and 24 (p=0.001), with both MET (p=0.005 and p=0.002, respectively) and SWITCH (p=0.002 at both times) superior to CONTROL, with effect sizes from 0.40 to 0.81 and no significant difference between MET and SWITCH (Table 2 and Figure 1).

Figure 1.

Estimated change in body mass index (BMI) z‐score over time

Secondary anthropometric outcomes

All other anthropometric measures followed the same pattern as the primary outcome (Table 2). Weight increased significantly in CONTROL (week 12: +4.3±0.8 lbs, p<0.0001; week 24: +8.5±1.5 lbs, p<0.001), while remaining essentially the same in MET (+0.3±0.7 lbs and –0.4±1.4 lbs) and SWITCH (+0.3±0.9 lbs and +0.3±1.9 lbs). The 3‐way comparison was significant at week 12 (p=0.0002) and 24 (p<0.0001), with both MET and SWITCH outperforming CONTROL at both times (effect size from 0.66 to 0.99).

Weight loss occurred in 55.1% of MET, 46.7% of SWITCH and 10.6% of CONTROL. Furthermore, 88.6% of CONTROL subjects gained weight, with 22.7% gaining ≥7% of baseline weight, compared to only 6.1% of MET and 9.7% of SWITCH.

Metabolic parameters

Of all measured metabolic parameters, only fasting glucose at week 12 differed significantly in the 3‐way comparison (p=0.001), with MET having a large effect (p=0.001, effect size 0.95) and SWITCH a medium effect (p=0.036, effect size 0.62) vs. CONTROL. MET also showed significant reductions in insulin (p=0.002) and HOMA‐IR (p=0.0002) at week 12, and HgbA1c (p=0.028) at week 24. No group showed significant changes in lipids or C‐reactive protein.

Psychiatric symptoms

Psychiatric symptoms, measured by the blindly‐rated BPRS‐C, improved in all groups and did not differ between groups (at week 24: CONTROL –4.3±1.7, MET –4.6±1.6, SWITCH –4.8±2.2). About 1/5 of each group showed much or greater improvement on the blindly‐rated CGI‐I. Deterioration (CGI‐I ≥5) occurred in about 10% of CONTROL and MET, none of aripiprazole‐SWITCH, 52.9% of perphenazine‐SWITCH and the only molindone‐SWITCH participant.

Weight‐related quality of life improved in all groups, with the only between‐treatment differences seen in the social life subscale at week 12 (p=0.005), favoring MET over SWITCH (p=0.001), and the physical comfort subscale at week 24 (p=0.026), favoring SWITCH over CONTROL (p=0.008).

Adverse events

Table 3 summarizes adverse events with overall treatment group differences of p<0.10. Discontinuations due to adverse effects were infrequent (3/47 in CONTROL, 4/46 in MET, and 1/30 in SWITCH).

Table 3.

Adverse events with overall treatment group differences (adverse event population)

	Control (N=45)	Metformin (N=47)	Switch (N=30)	p (overall)	p (MET vs. others)
Metformin > Others (%)
Decreased appetite, all severities	17.8	51.1	30.0	0.0028	0.0016
Diarrhea, all severities	22.2	51.1	23.3	0.0052	0.0016
Abdominal pain or discomfort, all severities	20.0	42.5	16.7	0.015	0.0066
Abdominal pain or discomfort, moderate/severe	0	10.6	0	0.026	0.0074
Infection, all severities	42.2	63.8	36.7	0.037	0.015
Vomiting or nausea, moderate/severe	0	8.5	0	0.040	0.020
Cough, all severities	17.8	34.0	13.3	0.068	0.027
Encopresis, all severities	0	10.6	3.3	0.071	0.031
Metformin < Others (%)
Aggression or hostility, all severities	31.1	6.4	2.3	0.0085	0.0043
Aggression or hostility, moderate/severe	22.2	6.4	20.0	0.078	0.038
Anger or irritability, all severities	31.1	10.6	36.7	0.014	0.0049
Impulse control disorder, all severities	15.5	2.1	20.0	0.028	0.017
Control > Others (%)
Abnormal weight gain, all severities	17.8	2.1	3.3	0.010	0.087
Paresthesia, all severities	6.7	0	0	0.062	NS
Restlessness, moderate/severe	6.7	0	0	0.062	NS
Switch > Others (%)
Dystonia, all severities	0	2.1	13.3	0.019	NS
Energy increased, moderate/severe	0	0	6.7	0.059	NS
Obsessive rumination, moderate/severe	0	0	6.7	0.059	NS
Hypoacusis, all severities	4.4	0	10.0	0.067	NS

Metformin was associated with significantly more abdominal pain, both moderate/severe (p=0.0074) and all severities (p=0.0066); infection, all severities (p=0.015); decreased appetite, all severities (p=0.0016); diarrhea, all severities (p=0.0016); vomiting or nausea, moderate/severe (p=0.020); and encopresis, all severities (p=0.031).

Perphenazine had numerically, but not statistically, higher proportions of both moderate/severe and all severities of hypersomnia and all severities of initial insomnia. Aripiprazole had statistically higher rates of mild dystonia (p=0.013).

MET was associated with significantly fewer problems with aggression/hostility, both moderate/severe (p=0.038) and all severities (p=0.0043), and had fewer all severity reports of anger/irritability (p=0.0049) and impulsiveness (p=0.017).

DISCUSSION

This is the first randomized trial to directly compare multiple strategies (behavioral HLE, add‐on metformin, or switch to a lower weight gain‐risk antipsychotic) for reducing antipsychotic‐associated weight gain in youth. Weight‐related outcomes for both MET and SWITCH were superior to CONTROL, with effect sizes ranging from 0.40 to 0.99, without being different from one another.

Since the weight reduction slopes in the two active arms appear to be linear, without reaching any plateau during the 6‐month study period, longer studies are needed to confirm that benefits will continue to grow with time, and to determine when these benefits start to plateau.

Metabolic benefits and inflammatory changes were minimal and seemed to diminish over time. Most participants remained psychiatrically stable, except in the perphenazine‐SWITCH arm, which was closed due to frequent psychiatric exacerbations resulting in discontinuation of 35.2% of subjects. Overall, both MET and aripiprazole‐SWITCH were well tolerated, although MET was associated with significantly more gastrointestinal adverse effects.

Despite these adverse events, MET was not associated with significantly greater treatment discontinuation for adverse effects. This result supports the observation that most gastrointestinal adverse effects were mild to moderate, occurred early during the titration phase, and were mostly transient, or could be managed by slowing down the dose titration and/or remaining at a lower dose of metformin.

Conversely and surprisingly, MET was associated with significantly fewer reports of problems with aggression/hostility and impulsiveness than CONTROL or SWITCH. It remains unclear whether this might relate to fewer food‐related struggles due to reduced appetite and/or MET's actions on glucose homeostasis in the brain or on cognition, similar to those observed in animal models43, 44.

The cardiometabolic and adverse effect results are generally consistent with those of three smaller and shorter metformin studies in antipsychotic‐treated youth, demonstrating cessation of ongoing weight gain, but minimal weight loss with metformin and minimal metabolic effects over the study duration. However, this study expands upon all existing metformin studies in antipsychotic‐treated individuals by also demonstrating comparable weight benefits with switch to a lower risk antipsychotic.

The minimal metabolic benefits of metformin are consistent with prior studies, and may be due to the fact that patients were selected for prior body weight gain, not metabolic abnormality. Nevertheless, individual group findings diverged somewhat from studies in adults16, 17, 18, 19, 34, 45, 46, in that neither MET nor SWITCH were associated with weight loss and the HLE behavioral intervention was associated with ongoing weight gain. This difference may be related both to normal developmental mechanisms promoting ongoing growth in youth, and the prolonged antipsychotic exposure of most participants.

Additionally, relative to the physiological growth taking place during the 6‐month study period, youth in the MET and SWITCH groups had negative sex‐ and age‐adjusted BMI z‐score and BMI percentile changes, whereas youth in the CONTROL group experienced not only increased body weight but also increases in BMI z‐score and BMI percentile values. This result indicates that, relative to normal development, MET and SWITCH led to a reduction in body weight, whereas CONTROL was associated with weight gain in addition to what would be expected during growth. Importantly, the differences between CONTROL and MET or SWITCH increased between week 12 (effect sizes 0.67 and 0.66) and week 24 (effect sizes 0.99 and 0.91), suggesting that continued use of MET or SWITCH likely increases the benefit.

The high rate of psychiatric destabilization with perphenazine based on blinded CGI and BPRS‐C assessments was unexpected, given that this drug had comparable efficacy to multiple SGAs in adults with schizophrenia33. However, its efficacy for pediatric psychotic and mood disorders has never been evaluated. The observed destabilization does not appear to be the result of too rapid a switch (given the very slow plateau cross‐titration and the fact that 41.2% failed to discontinue their baseline antipsychotic) or extrapyramidal adverse effects (given the use of prophylactic anticholinergic treatment and only mild parkinsonian symptoms in two participants). However, adverse events limited our ability to increase the perphenazine dose as much as desired, suggesting a reduced benefit‐risk ratio of perphenazine in youth.

The results of this study need to be interpreted within its limitations. First, although this was the largest metformin study conducted to date in antipsychotic‐treated youth, individual group sizes were still modest, especially in the SWITCH arm, and secondary analyses were not corrected for multiple testing. Metabolic changes might be more evident in a larger sample.

Second, there was a smaller number of participants in the SWITCH arm. Reasons included the halted randomization from when molindone was discontinued in the US market until approval of the perphenazine‐SWITCH arm, and later stoppage of switch to perphenazine due to increased psychiatric worsening. Since perphenazine‐SWITCH was not well tolerated and led to substantial discontinuation due to inefficacy, additional agents with low weight gain potential need to be studied. In this context, it remains unclear if switching from one lower risk agent, like aripiprazole, to another “lower risk” agent, including ziprasidone and lurasidone, or (based on adult data) cariprazine or brexpiprazole, would have similar effects to SWITCH in this study5, 47, 48.

Third, we did not compare MET or SWITCH to a formal weight loss intervention, but rather used HLE, consisting of education and close weight monitoring. Future studies are needed to investigate the efficacy of a formal weight loss intervention in overweight/obese, antipsychotic‐treated youth. Although prior studies suggested that formal weight loss interventions were efficacious in adults with antipsychotic‐induced weight gain18, the largest study of a behavioral weight loss intervention in antipsychotic‐treated adults failed to demonstrate its superiority to treatment‐as‐usual22.

Fourth, this was an open study without placebo control, likely influencing some participants to withdraw immediately post‐randomization. However, open treatment increased the ecologic validity of our findings, the majority of outcomes were objective measurements, and the BPRS‐C and CGI were assessed by blinded evaluators, minimizing risk of bias.

Fifth, we decided on a relatively slow metformin titration over 4 weeks, in order to minimize dose‐ and titration‐dependent adverse effects that could have increased undesirable drop‐out rates. While it is possible that a faster metformin titration could have yielded larger effects, the rather linear slope of the BMI z‐score change does not suggest major acceleration of the efficacy as higher doses were achieved.

Sixth, patients could be on a total of three (or four, if two medications were used for ADHD) psychotropic medications. While co‐medication effects could potentially have confounded the results, this study methodology assured higher generalizability, as patients receiving antipsychotics often are on multiple psychotropic drugs. Further, the randomized groups differed by only 3‐5% regarding use of psychostimulants and antidepressants and 8‐14% concerning use of mood stabilizers.

Finally, while the rate of concomitant psychostimulant treatment was relatively high across the three intervention groups (40‐45%), this unlikely affected the results. Prior studies have shown that antipsychotic‐related weight gain is not moderated or diminished by concomitant psychostimulant use49, 50. Moreover, patients in this study all had significant weight gain despite the fact that they received concomitant psychostimulant treatment that was kept stable during the study.

In summary, our results provide evidence that both MET and switch to aripiprazole reduce SGA‐related weight gain burden in youth. Further study is required to determine whether, as suggested in adults21, the benefits of MET might be stronger in youth with more limited weight gain or during antipsychotic initiation51. However, the minimal effect observed on metabolic outcomes mandates routine metabolic monitoring and careful consideration of other treatment strategies with lower risk for weight gain prior to the antipsychotic initiation52.

Further, although aripiprazole switch was associated with significant reductions in weight measures vs. the control group, the high prevalence of its current (46.4%) and prior (26.0%) use in this sample suggests that a limited subset of children may benefit from this strategy. Evaluation of other potential switch agents, metformin's prophylactic use, other agents that may mitigate antipsychotic‐associated weight gain, and the potential benefit of MET for aggression is warranted in youth.